5.5 Experiment Six: Selecting a feature pair
5.5.1 Rationale of Experiment Six
Before embarking on a study to explore the effects of a top-down set from the SVT on change detection performance, it was important to ensure that a change could be detected without the addition of the SVT, and that any change was not too easy to detect as this may undermine any effects of the carry-over. A selection of features was
therefore piloted in a change detection task in order to select two that would be used for the future experiments.
Although the SVT is not an example of the most basic visual search task (search for a feature singleton), the search features were selected on the assumption that they were ‘basic features’ (Wolfe, 1994) and attention could be allocated to these features selectively. The intention was to use two search features, therefore in the first pilot study three feature pairs were chosen; colour and shape, height and width, and spatial frequency and orientation. It is generally accepted that colour is a basic search feature and studies show that observers can selectively attend to a subset of colour (e.g., Egeth et al., 1984; Kaptein et al., 1995). Wolfe also suggests that orientation and size are basic search features. Size was manipulated by altering the height of an item, or the width of an item. Orientation was paired with spatial frequency, using grating patterns with different spatial frequencies, with the gratings slanting to the left or the right. Wolfe suggests that spatial frequency is similar to size in basic visual search, and cites work by Sagi (1988) which agrees with this. In addition, previous research has shown that the spatial frequency and orientation of centrally located grating patterns can influence detection of peripherally located grating patterns (Rossi & Paradiso, 1995). Specifically, when a participant is asked to detect the orientation of a central grating pattern they are more sensitive to the same orientation of peripheral grating patterns, and when a participant is asked to respond to the spatial frequency of a central grating they are more sensitive to similar spatial frequencies of peripheral gratings. However, Rossi and Paradiso also found that when responding to spatial frequency, the orientation of the central grating could not be ignored and it also enhanced performance for orientation search in the peripheral gratings.
As there was a different pattern of performance for orientation and spatial frequency it indicates that spatial frequency may not be a basic feature. However, given the success of this previous study in showing selective attention to features, and transference of this selective attention to additional visual images, the feature pairing was retained for the pilot study.
In his 1994 paper Wolfe does not describe shape as a basic feature, however he later described it as a “probable guiding attribute” (Wolfe & Horowitz, 2004) illustrating that there is uncertainty regarding whether it can be defined as a basic feature. However, like the findings of spatial frequency reported by Rossi and Paradiso (1995), Ghirardelli and Egeth (1998) found that shape could guide visual attention. If participants were told the shape of a target before completing a search task only distracters matching that shape interfered with target detection. Yet when participants were not told which shape the target would be before the search array was presented they suffered interference from all distracters equally. As in the previous instance, shape was retained as a feature because there is evidence to suggest that it can influence the allocation of attention, despite the fact that it may not be a basic feature.
Change detection performance was therefore measured for six features (making up three feature pairs). The ideal feature pair would be one in which change detection performance was relatively equal for both features, and that the change was fairly difficult to detect (to allow for possible benefits of a congruent SVT to be determined). Participants completed three blocks of trials, one for each feature pair.
5.5.2 Method
5.5.2.1 Participants:
Sixteen participants completed the experiment, seven males and nine females. All were aged between 19 and 28 with a mean age of 21.8 years. All reported normal or corrected-to-normal vision. Participants were a mixture of undergraduate and postgraduate students at the University of Nottingham.
5.5.2.2 Design
The study used a within-participants design with two variables, the first was
feature pair, and the second was feature. Feature pair represented the three pairs of
features which were combined in each block; these were height and width, colour and shape, and spatial frequency and orientation. Feature referred to the two different features within each feature pair. The dependent variables were accuracy in locating the changing item, and RT to detect the change.
5.5.2.3 Apparatus and Stimuli:
A total of 384 visual arrays were used for the experiment; 192 original arrays and 192 modified arrays. Each modified array was identical to an original array with a single item changed. The arrays consisted of 16 shapes on a white background, placed at one of 25 locations within an invisible 5x5 grid. Each array measured 20.8°, and each individual item measured a maximum of 3.3°. In the height-width block the shapes were black rectangles and were either ‘short and wide’, ‘tall and wide’, ‘short and narrow’, or ‘tall and narrow’. In the colour-shape block the shapes were a mixture of diamonds and triangles and were coloured blue or purple with a black outline of 2mm. In the spatial frequency-orientation block the shapes were all circles filled with
black and white bars (similar to a sinusoidal grating pattern); spatial frequency was either low (1.5 cycles per degree) or high (3 cycles per degree) and the bars were either slanting 30º to the left or 30º to the right.
Items were placed within the grid in a pseudorandom manner. Locations (1- 25) were assigned to each item for each array; items in an already occupied location were moved to the nearest available location and 9 locations were left unoccupied. There were never more than 5 or less than 3 identical items in any array. The type of change was pre-selected at random, ensuring that there were an equal number of changes made to each of the four different items within each block. See figure 5.1 (page 161) for an example of the images used in the experiment.
5.5.2.4 Procedure:
The 192 trials each began with a black fixation cross shown to the centre of the screen for 500ms. After this an original array was shown for 3000ms11. A blue blank screen then replaced the original array for 200ms and following this the original array was shown for a further 500ms. A blue blank screen was again shown for 200ms before the matching modified array was presented for 500ms. The two arrays then continued to alternate (separated by the blue blank screen and presented for 500ms each time), until participants had successfully located the change in the array. Once the change had been found participants were told to press the spacebar, they then saw a response screen separated into five numbered sections (these sections corresponded to the horizontal rows of stimuli in the arrays). They had to press numbers 1-5 to state which row the changing shape had been located. After making a response feedback
11 The original array was shown initially to represent the display of an array in the SVT task. In the
pilot studies participants were simply told to view this array and no response was required, however it was important to set up the study in a similar way to later experiments to ensure that any carry-over effect was due to the search required in the SVT, and that additional effects of the display were controlled for in this baseline measure of change detection performance.
was provided for 400ms before the next trial began. See figure 5.2 on page 163 for the sequence of events in every trial.
Figure 5.1: Examples of the original and modified arrays used in the experiment. The uppermost arrays
were used in the height-width condition, here the change is to the first item on the fourth row and is a height change. In the colour-shape arrays the change has been made to the last item on the second row and this is a shape change. In the spatial frequency-orientation arrays the first circle on the fourth row has changed its orientation, with the bars moving from right to left.
Trials were separated into three blocks (colour and shape, height and width, and spatial frequency and orientation) and the order of these blocks was randomised
across participants. This allowed 64 trials for each feature pair. There were four different items in each array and each item could change in two different ways (see table 5.1), in 32 of these trials items changed by feature 1 (e.g., colour), in the other 32 trials items changed by feature 2 (e.g., shape). As there were a total of 8 possible changes in each array, each change-type occurred 8 times in each block. All trials were presented randomly.
Block Original Change (feature 1) Change (feature 2)
Height-Width Short and wide Tall and wide (H) Short and narrow (W) Tall and wide Short and wide (H) Tall and narrow (W) Short and narrow Tall and narrow (H) Short and wide (W) Tall and narrow Short and narrow (H) Tall and wide (W) Colour-Shape Blue diamond Purple diamond (C) Blue triangle (S)
Purple diamond Blue diamond (C) Purple triangle (S) Blue triangle Purple triangle (C) Blue diamond (S) Purple triangle Blue triangle (C) Purple diamond (S) Spatial frequency-
Orientation Low SF left High SF left (SF) Low SF right (O) High SF left Low SF left (SF) High SF right (O) Low SF right High SF right (SF) Low SF left (O) High SF right Low SF right (SF) High SF left (O)
Table 5.1: The possible changes that could be made to items in the experiment separated into the three
feature pairs. Changes made to feature 1 were height (H), colour (C), and spatial frequency (SF),
Figure 5.2: The temporal sequence of events in Experiment Six.
Arrays continue to alternate until the change has been found
Fixation (500ms)
Original array (3000ms)
Blue blank screen (200ms)
Blue blank screen (200ms) Original array (500ms)
Modified array (500ms)
Blue blank screen (200ms)
Original array (500ms)
Response screen
5.5.3 Results
Analysis consisted of two 1x3 within-participants ANOVAs which compared accuracy and RT between the three feature pairs. Paired samples t-tests were then carried out to compare the two features within each feature pair. Prior to the analysis any responses made before 800ms12 and after 30 seconds were removed from the data. In addition, a correlation was conducted for each condition to check for any speed- accuracy trade-off. This showed no relationship between accuracy and speed, indicating that participants were not completing the experiment using a specific strategy whereby they place more emphasis on accuracy (and are therefore slower) or speed (and are therefore less accurate).13
For change detection accuracy there was no significant difference between the three feature pairs, and performance was reaching an average of 98%. This level of accuracy is standard for a change detection task, and accuracy is not often used as a dependent measure in most flicker paradigms. Incorrect trials were removed from the analysis at this stage, following the practice of previous flicker experiments cited in the literature. For RT to correctly detected changes there was a main effect of feature
pair (F (2,45) = 31.570, MSE = 2.213, p<0.001). Post-hoc comparisons using the
Bonferroni correction showed that RT in the spatial frequency-orientation block (¯ = 9.43 seconds) was significantly longer than RT in the other two blocks
x
(p<0.016). RT in the height-width block (
x
¯ = 5.47 seconds) did not differ from RT in the colour-shape block (x
¯ = 6.3 seconds). There was no significant difference between the two features within each feature pair with regard to accuracy, and there was no12 The modified array did not appear on the screen until 700ms after the first presentation of the
original array and the blue blank screen, therefore a change could not have been detected before 800ms.
13 These steps were taken for all subsequent experiments using the visual search-change detection
methodology, but will only be mentioned again in instances where outliers (or a relationship between speed and accuracy) were found in each experiment.
difference in RT between height and width. Participants were however detecting a colour change faster than a shape change (t (15) = -2.355, p<0.05), and were detecting a change to spatial frequency faster than a change to orientation (t (15) = -4.732, p<0.001). See figure 5.3 for these findings.
0 2 4 6 8 10 12 HW CS SFO C ha ng e de te ct io n R T (s ec s)
Figure 5.3: Mean change detection response times for each feature within each feature pair, height and
width (HW), colour and shape (CS), and spatial frequency and orientation (SFO).
5.5.4 Discussion
The aim of the pilot study was to measure change detection of six features separated into three feature pairs. This was in order to select a feature pair to use for subsequent experiments in this area. Change detection had to be sufficiently difficult to warrant focused attention, and to ensure that any carry-over from the SVT could be measured. Three feature pairs were contrasted and findings showed that changes to height and width were possibly too easy to detect and changes to spatial frequency and orientation were possibly too difficult to detect. Whilst this did not reveal itself in the accuracy data, changes to spatial frequency and orientation took significantly
longer to detect than changes made to the other two pairs of features. Although it was necessary to use features which required focused attention, if the changes take a long time to find it will prolong the experiment, reducing the number of trials which can be completed. The pairing of colour and shape was therefore selected as the most
suitable for the subsequent experiments.
Aside from determining the most appropriate feature pair to use for later experiments, the pilot study did raise one interesting finding which was that an orientation change was more difficult to detect than a spatial frequency change. According to Wolfe (1994) orientation is a basic feature, and Rossi and Paradiso (1995) have demonstrated that whilst participants can selectively attend to spatial frequency, attention is allocated to orientation regardless of the task demands. In light of this it would be expected that an orientation change would be easier to detect. The present finding may have been due to the stimuli used in the experiment, as it is difficult to quantify if changes to orientation were as substantial as changes made to spatial frequency. Yet it implies that orientation may only be defined as a basic feature when it is presented as a single feature, and not part of a feature conjunction. To illustrate, in visual search studies cited in the literature, a search for orientation involves a search for a single line which is slanted to the left or right. In the present search task orientation was part of an object that could be defined on the basis of two separate features, or a conjunction of features. In this latter instance it may be more difficult to search selectively. This suggestion would of course require additional investigation, and whilst the finding does have implications regarding the
requirements to be defined as a ‘basic feature’, these implications are not directly related to the focus of this thesis.